2.3.1 Summary stats checks

qdap.reviews2[, c('incentivized','ave.polarity')] %>%
  group_by(incentivized) %>%
  summarize_all(mean, na.rm = TRUE)

## # A tibble: 2 x 2
##   incentivized     ave.polarity
##   <fct>                   <dbl>
## 1 incentivized            0.277
## 2 non-incentivized        0.406

This result is different from the one we saw using ‘sentimentr’ package

## check the records with na values for the ave.polarity
knitr::kable(qdap.nas <- qdap.reviews2[is.na(qdap.reviews2$ave.polarity),],
             caption = 'All reviews with NA ave.polarity score', floating.environment="sidewaystable")

There’re some NA values for edge cases

2.3.2 Plots

Boxplot: rating vs. ave.polarity

## Warning: Ignoring 68 observations

2.3.3 Fixed Effect Linear Model

Join with selected columns from the raw data

qdap.reviews2 <- merge(qdap.reviews2, reviews2.csv[,c('recid','item_id')])

## fix effect linear model
## Use sentence sentiment score to replce rating
formula.fe <- ave.polarity ~ incentivized + is_deleted + verified_purchaser
model.fe <- plm(data = qdap.reviews2, formula = formula.fe, index = c('item_id'), model = 'within')
# get the model summary
summary(model.fe)

## Oneway (individual) effect Within Model
## 
## Call:
## plm(formula = formula.fe, data = qdap.reviews2, model = "within", 
##     index = c("item_id"))
## 
## Unbalanced Panel: n = 101, T = 29-10133, N = 263948
## 
## Residuals:
##      Min.   1st Qu.    Median   3rd Qu.      Max. 
## -2.762940 -0.300630 -0.058203  0.237060  4.102307 
## 
## Coefficients:
##                                Estimate Std. Error t-value  Pr(>|t|)    
## incentivizednon-incentivized  0.0746480  0.0113962  6.5502 5.755e-11 ***
## is_deletedkept               -0.0282067  0.0037962 -7.4303 1.087e-13 ***
## verified_purchaserverified    0.0397443  0.0038586 10.3002 < 2.2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Total Sum of Squares:    54423
## Residual Sum of Squares: 54379
## R-Squared:      0.00080966
## Adj. R-Squared: 0.00041959
## F-statistic: 71.2656 on 3 and 263844 DF, p-value: < 2.22e-16

2.3.4 Correlation: rating vs. review sentiment

cor.test(qdap.reviews2$rating, qdap.reviews2$ave.polarity, method=c("pearson", "kendall", "spearman"))

## 
##  Pearson's product-moment correlation
## 
## data:  qdap.reviews2$rating and qdap.reviews2$ave.polarity
## t = 280.03, df = 263946, p-value < 2.2e-16
## alternative hypothesis: true correlation is not equal to 0
## 95 percent confidence interval:
##  0.4756451 0.4815274
## sample estimates:
##       cor 
## 0.4785916

3.2.1 Plots

Boxplot: rating vs. syuzhet.sentiment

Boxplot: rating vs. afinn.sentiment

Boxplot: rating vs. nrc.sentiment

Boxplot: rating vs. bing.sentiment

3.2.2 Fixed Effect Linear Model

Join with selected columns from the raw data

syuzhet.reviews2 <- merge(syuzhet.reviews2, reviews2.csv[,c('recid','item_id')])

Here we’re showing the linear model for all four options

## fix effect linear model
## Use sentence sentiment score to replce rating
formula.fe <- syuzhet.sentiment ~ incentivized + is_deleted + verified_purchaser
model.fe <- plm(data = syuzhet.reviews2, formula = formula.fe, index = c('item_id'), model = 'within')
# get the model summary
summary(model.fe)

## Oneway (individual) effect Within Model
## 
## Call:
## plm(formula = formula.fe, data = syuzhet.reviews2, model = "within", 
##     index = c("item_id"))
## 
## Unbalanced Panel: n = 101, T = 29-10134, N = 264016
## 
## Residuals:
##      Min.   1st Qu.    Median   3rd Qu.      Max. 
## -10.05502  -0.77039  -0.19341   0.58073  28.34164 
## 
## Coefficients:
##                               Estimate Std. Error  t-value  Pr(>|t|)    
## incentivizednon-incentivized -4.894152   0.035167 -139.169 < 2.2e-16 ***
## is_deletedkept               -0.230489   0.011713  -19.678 < 2.2e-16 ***
## verified_purchaserverified   -0.403659   0.011906  -33.904 < 2.2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Total Sum of Squares:    574120
## Residual Sum of Squares: 517970
## R-Squared:      0.097792
## Adj. R-Squared: 0.09744
## F-statistic: 9535.34 on 3 and 263912 DF, p-value: < 2.22e-16

## fix effect linear model
## Use sentence sentiment score to replce rating
formula.fe <- afinn.sentiment ~ incentivized + is_deleted + verified_purchaser
model.fe <- plm(data = syuzhet.reviews2, formula = formula.fe, index = c('item_id'), model = 'within')
# get the model summary
summary(model.fe)

## Oneway (individual) effect Within Model
## 
## Call:
## plm(formula = formula.fe, data = syuzhet.reviews2, model = "within", 
##     index = c("item_id"))
## 
## Unbalanced Panel: n = 101, T = 29-10134, N = 264016
## 
## Residuals:
##       Min.    1st Qu.     Median    3rd Qu.       Max. 
## -30.112833  -2.315512  -0.037194   1.701766  52.469213 
## 
## Coefficients:
##                               Estimate Std. Error t-value  Pr(>|t|)    
## incentivizednon-incentivized -8.441802   0.089407 -94.419 < 2.2e-16 ***
## is_deletedkept               -0.511688   0.029778 -17.183 < 2.2e-16 ***
## verified_purchaserverified   -0.686430   0.030270 -22.677 < 2.2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Total Sum of Squares:    3519300
## Residual Sum of Squares: 3348000
## R-Squared:      0.048664
## Adj. R-Squared: 0.048293
## F-statistic: 4500.04 on 3 and 263912 DF, p-value: < 2.22e-16

## fix effect linear model
## Use sentence sentiment score to replce rating
formula.fe <- nrc.sentiment ~ incentivized + is_deleted + verified_purchaser
model.fe <- plm(data = syuzhet.reviews2, formula = formula.fe, index = c('item_id'), model = 'within')
# get the model summary
summary(model.fe)

## Oneway (individual) effect Within Model
## 
## Call:
## plm(formula = formula.fe, data = syuzhet.reviews2, model = "within", 
##     index = c("item_id"))
## 
## Unbalanced Panel: n = 101, T = 29-10134, N = 264016
## 
## Residuals:
##       Min.    1st Qu.     Median    3rd Qu.       Max. 
## -11.143271  -0.925852  -0.080661   0.587376  32.544075 
## 
## Coefficients:
##                               Estimate Std. Error  t-value  Pr(>|t|)    
## incentivizednon-incentivized -5.320651   0.040027 -132.928 < 2.2e-16 ***
## is_deletedkept               -0.195991   0.013331  -14.701 < 2.2e-16 ***
## verified_purchaserverified   -0.356488   0.013551  -26.306 < 2.2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Total Sum of Squares:    733780
## Residual Sum of Squares: 671020
## R-Squared:      0.085524
## Adj. R-Squared: 0.085167
## F-statistic: 8227.26 on 3 and 263912 DF, p-value: < 2.22e-16

## fix effect linear model
## Use sentence sentiment score to replce rating
formula.fe <- bing.sentiment ~ incentivized + is_deleted + verified_purchaser
model.fe <- plm(data = syuzhet.reviews2, formula = formula.fe, index = c('item_id'), model = 'within')
# get the model summary
summary(model.fe)

## Oneway (individual) effect Within Model
## 
## Call:
## plm(formula = formula.fe, data = syuzhet.reviews2, model = "within", 
##     index = c("item_id"))
## 
## Unbalanced Panel: n = 101, T = 29-10134, N = 264016
## 
## Residuals:
##      Min.   1st Qu.    Median   3rd Qu.      Max. 
## -12.40467  -1.01461  -0.15270   0.87808  29.59394 
## 
## Coefficients:
##                               Estimate Std. Error  t-value  Pr(>|t|)    
## incentivizednon-incentivized -4.904944   0.046670 -105.097 < 2.2e-16 ***
## is_deletedkept               -0.299480   0.015544  -19.266 < 2.2e-16 ***
## verified_purchaserverified   -0.377601   0.015801  -23.898 < 2.2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Total Sum of Squares:    969570
## Residual Sum of Squares: 912270
## R-Squared:      0.0591
## Adj. R-Squared: 0.058733
## F-statistic: 5525.68 on 3 and 263912 DF, p-value: < 2.22e-16

3.2.3 Correlation: rating vs. review sentiment

cor.test(syuzhet.reviews2$rating, syuzhet.reviews2$syuzhet.sentiment, method=c("pearson", "kendall", "spearman"))

## 
##  Pearson's product-moment correlation
## 
## data:  syuzhet.reviews2$rating and syuzhet.reviews2$syuzhet.sentiment
## t = 143.17, df = 264014, p-value < 2.2e-16
## alternative hypothesis: true correlation is not equal to 0
## 95 percent confidence interval:
##  0.2648695 0.2719488
## sample estimates:
##       cor 
## 0.2684128

cor.test(syuzhet.reviews2$rating, syuzhet.reviews2$afinn.sentiment, method=c("pearson", "kendall", "spearman"))

## 
##  Pearson's product-moment correlation
## 
## data:  syuzhet.reviews2$rating and syuzhet.reviews2$afinn.sentiment
## t = 183.08, df = 264014, p-value < 2.2e-16
## alternative hypothesis: true correlation is not equal to 0
## 95 percent confidence interval:
##  0.3322494 0.3390189
## sample estimates:
##       cor 
## 0.3356385

cor.test(syuzhet.reviews2$rating, syuzhet.reviews2$nrc.sentiment, method=c("pearson", "kendall", "spearman"))

## 
##  Pearson's product-moment correlation
## 
## data:  syuzhet.reviews2$rating and syuzhet.reviews2$nrc.sentiment
## t = 90.225, df = 264014, p-value < 2.2e-16
## alternative hypothesis: true correlation is not equal to 0
## 95 percent confidence interval:
##  0.1692463 0.1766471
## sample estimates:
##       cor 
## 0.1729491

cor.test(syuzhet.reviews2$rating, syuzhet.reviews2$bing.sentiment, method=c("pearson", "kendall", "spearman"))

## 
##  Pearson's product-moment correlation
## 
## data:  syuzhet.reviews2$rating and syuzhet.reviews2$bing.sentiment
## t = 196.88, df = 264014, p-value < 2.2e-16
## alternative hypothesis: true correlation is not equal to 0
## 95 percent confidence interval:
##  0.3544634 0.3611157
## sample estimates:
##       cor 
## 0.3577941